Circulating cell-free RNA in blood as a host response biomarker for detection of tuberculosis

Abstract

Tuberculosis (TB) remains a leading cause of death from an infectious disease worldwide, partly due to a lack of effective strategies to screen and triage individuals with potential TB. Whole blood RNA signatures have been tested as biomarkers for TB, but have failed to meet the World Health Organization's (WHO) optimal target product profiles (TPP). Here, we use RNA sequencing and machine-learning to investigate the utility of plasma cell-free RNA (cfRNA) as a host-response biomarker for TB in cohorts from Uganda, Vietnam and Philippines. We report a 6-gene cfRNA signature, which differentiates TB-positive and TB-negative individuals with AUC = 0.95, 0.92, and 0.95 in test, training and validation, respectively. This signature meets WHO TPPs (sensitivity: 97.1% [95% CI: 80.9-100%], specificity: 85.2% [95% CI: 72.4-100%]) regardless of geographic location, sample collection method and HIV status. Overall, our results identify plasma cfRNA as a promising host response biomarker to diagnose TB.

Fig. 1. Plasma cell-free RNA profiling.

A Geographic distribution of samples included in this study, which originated from three different cohorts. B Differences in cfRNA cell-types-of-origin between the three cohorts. Light-blue = Platelet, red = Myeloid-progenitor, yellow = B cell, dark-green = Endothelial cell, maroon = NK cell, dark-blue = Monocyte, light-green = Neutrophil, beige = Other. C VST normalized counts of significantly differentially abundant genes (two sided Wald test, Benjamini-Hochberg adjusted p-value < 0.05). Samples and genes are clustered based on correlation. Country: dark-green = Uganda, yellow = Vietnam, red = Philippines; Cohort: light-blue = Cohort 1, purple = Cohort 2, gold = Cohort 3; HIV status: orange = positive, maroon = negative; TB status: pink = positive, dark-blue = negative. D Top 21 differential pathways between microbiologically confirmed TB diagnoses ranked by significance. Z-score indicated in corresponding bar; direction relative to TB negative samples. E Flowchart of the method used to train, test, and validate the machine learning classification algorithms. F Area under the receiver operating characteristic curve (ROC-AUC) metrics for training and test sets across 15 machine learning models (test set = triangle, training set = circle).

Fig. 2. Performance of the 6-gene signature identified in plasma cfRNA.

A Area under the receiver operating characteristic curve (ROC-AUC) as a function of the gene is added in each iteration of a greedy forward search model performed on the training dataset. B Train (dark-green), test (purple), and validation (gold) performance of the greedy forward search algorithm in distinguishing microbiologically confirmed TB. C Violin plot of classifier scores for training, test, and validation sets using the greedy forward search algorithm (TB positive = pink, TB negative = blue; HIV positive = triangle, HIV negative = circle). The dashed line represents the optimal Youden threshold cutoff (0.498) which was determined solely on the training set. D VST normalized counts of the most significant predictor GBP5 across both cohort and semiquantitative TB status (blue = negative, dark-green = low, yellow = medium, red = high). Boxes in the boxplots indicate the 25th and 75th percentiles, the band in the box represents the median, and whiskers extend to 1.5 x interquartile range of the hinge. E Performance of the 6-gene TB score when the model is re-evaluated without HIV-positive individuals. F Performance of the 6-gene TB score when separating samples by country (red = Uganda, green = Vietnam, light-blue = Philippines). G Correlation of the 6-gene TB score with the Xpert Ultra Semi-quantitative Result (dotted line: classification score threshold = 0.498; center: mean; bars indicate 95% confidence interval +/− SEM). Light-blue = Cohort 1, brown = Cohort 2, gold = Cohort 3. H Correlation of the 6-gene TB score with three chest X-ray scores. Color indicates disease status (pink = TB positive; blue = TB negative) Error bands represent the 95% CI around the LOESS smoothed line.

Fig. 3. Comparison of the 6-gene plasma signature with whole blood signatures of TB from literature.

A Overlap between top-performing whole blood signatures (dark-blue) and the 6-gene signature (pink). Top 29 overlapping genes are shown. B Performance comparison of whole blood signatures from previous multi-cohort studies (dark-blue) with our 6-gene cfRNA signature (pink). Optimal triage thresholds are marked in light-blue (95% sensitivity, 80% specificity), minimal triage thresholds are marked in green (90% sensitivity, 70% specificity). Error bars represent the 95% confidence intervals. Samples sizes (n) are as follows: cfRNA (61), Berry86 (116), daCosta2 (54), daCosta3 (54), Kaforou44 (103), Walter47 (180), Zak16 (49), Sweeney3 (787). C Comparison of solid organ fraction between the 3 cfRNA cohorts (light-blue, brown, gold) and the whole blood samples (dark-blue). “****” represents a two-sided Wilcoxin test, Benjamini-Hochberg adjusted p-value < 1e-15. Boxes in the boxplots indicate the 25th and 75th percentiles, the band in the box represents the median, and whiskers extend to 1.5 × interquartile range of the hinge. D Venn diagram depicting the overlap of statistically significant, differentially expressed genes between whole blood and plasma cfRNA. E Performance of GBP5 cfRNA abundance in distinguishing active TB. F Performance comparison of whole blood protein GBP5, whole blood RNA GBP5, and plasma cfRNA GBP5 in distinguishing active TB.